Automatically version and trace llm.call methods with lilypad.configure

[willbakst]

We should make the following code snippet just work:

from mirascope import lilypad, llm

lilypad.configure()

@llm.call("openai", "gpt-4o-mini")
def recommend_book(genre: str) -> str:
    return f"Recommend a {genre} book"

response = recommend_book("fantasy")  # automatically versioned and traced
print(response.content)

The result should be the same as if using the lilypad.generation decorator except without it. This means the resulting closure we compute would be:

from mirascope import llm

@llm.call("openai", "gpt-4o-mini")
def recommend_book(genre: str) -> str:
    return f"Recommend a {genre} book"

We should also ensure that using lilypad.generation would still work, e.g.

from mirascope import lilypad, llm

@lilypad.generation()
@llm.call("openai", "gpt-4o-mini")
def recommend_book(genre: str) -> str:
    return f"Recommend a {genre} book"

Of course, this would include @lilypad.generation() in the computed closure unlike the first example where it's excluded.

We should keep this functionality in the lilypad branch until we're ready to release it.

[willbakst]

This will be further designed and implemented as part of Mirascope 2.0

For now I will close this issue and open a new issue in the future with more details when I have them.